2-osp-literature

/2-osp-literature — Literature Review & Expansion

Runs ONE round of external retrieval per invocation. Three rounds are recommended; the user decides how many to run. After each round it reports what it found and offers both routes: another round, or move on.

Activation

Invoke the osp-literature-review-agent skill.

Inputs — none of these is a gate

• phases.summary.status == "completed" and 01_structured_summary.md exists. Without it, read .brain/input/paper.md and derive the query terms yourself; say in Provenance that you did.
• Rounds run in order (1 → 2 → 3) because each strategy builds on the last. How many you run is the user's choice — one round is a valid, thinner corpus, not an error.

Resource notice

⚠️ Each invocation makes ~8-12 API calls across whichever databases this project installed, plus native web search. Expect 1-3 minutes per round.

Round definitions

#	Anchor	Goal
1	sub-domain-anchor	Search using the paper's stated sub-domain and primary keywords
2	method-anchor	Search using the method's name and key technical terms
3	temporal-expansion	Filter to last 12 months; include arXiv pre-prints, concurrent submissions

Opening block (print before the round starts)

Render the opening block exactly as .claude/defaults/phase_block_template.md defines it — that file holds the rail, the rules and the widths, and is the only place they are written down. This is phase 3 of 7 (literature); the rail carries ◐ while rounds remain. Values:

  DOING    <the anchor for this round, in one line>
  READS    .brain/raw/01_structured_summary.md
  WRITES   .brain/raw/02<a|b|c>_literature_round<N>.md
  COST     ~1-3 min, ~8-12 searches

Steps

1. Read .brain/session.json.
   • Determine next_round = phases.literature.rounds_completed + 1 (default 0 → next = 1).
   • If next_round > 3, all three recommended rounds are done: consolidate if that has not happened yet, and point at /3-osp-historian.
   • If any earlier round file is missing, resume from that round instead.